The most robust characterizations of paleoclimates are established when multiple proxies are present for the same climate characteristic. Where proxies disagree or multiple proxies are unavailable, as has been the case for Paleogene tropical SSTs, the climatic interpretation becomes muddied. As exemplified by the debate over tropical SSTs at the last glacial maximum, models or theory might shed some light, but only the collection of more proxy records is likely to resolve the issue. Nevertheless, great leaps in understanding occur only in situations in which theories and models fail to match key data. This requires theories and models that are independent of the data, to avoid circularity. This is the special purpose of having a null hypothesis and a straw man model.
In the past, it has been difficult to provide such a straw man because climate models have required either a specified SST or heat flux distribution as input. Thus, attempts to answer the question of what maintains the temperature and heat flux distributions become entangled in circular reasoning. In these models (summarized in DeConto et al., 2000), the surface freshwater balance is also not given a self-consistent treatment because realistic ocean-atmosphere interactions are not included. Uncoupled modeling efforts have been hampered by the fact that they must explicitly incorporate either assumptions based on modern-day relationships (e.g., modern ocean heat flux) or existing proxy-derived data (e.g., fixed SSTs) into the model because of the lack of interaction of the model components. Both approaches have deficiencies. The first ignores the potential for very different conditions in the past. The second raises issues of circularity in the experimental framework. In addition, the potential for an inconsistency between the ocean and atmospheric implied circulations is a fundamental challenge to uncoupled modeling techniques (Huber et al., 2001). Model simulations that do not treat atmosphere-ocean interaction or have a highly simplified atmosphere (or ocean) cannot adequately address the magnitude of the net surface energy and momentum fluxes because they do not properly represent changes in wind, temperature, precipitation fields, current structures, and modes of variability.
Although it has its own deficiencies, a fully coupled general circulation model (CGCM) permits us to avoid these pitfalls and provide a relatively independent model to compare climate proxy interpretations against. Fully coupled general circulation paleoclimate modeling is a new and rapidly evolving field, so it is important to understand the caveats. I discuss the difficulties and my methods in this paper, Huber and Sloan (2001), and Huber et al. (in press).
Most CGCMs require "flux corrections" to maintain a stable steady-state climate for modern conditions because of weaknesses in the model formulation. Without these corrections, the modeled climate drifts toward unrealistic values for important climate parameters. These models are not convergent on a realistic climate state even for perfectly known initial and boundary conditions and, consequently, cannot be used for paleoclimate modeling in the deep past. It is critical, therefore, to use a CGCM that does not require flux corrections for modern-day climate conditions when attempting to model the early Paleogene. The model used herein does not require flux corrections.
The model, the National Center for Atmospheric Research (NCAR) Climate System Model (CSM), is described below and by Boville and Gent (1998). The results presented here are produced with an updated version, 1.4, described for a modern and a Cretaceous experiment in Boville et al. (2001), Otto-Bliesner and Brady (2001), Otto-Bliesner et al. (2002), Large et al. (2001), and briefly below. The atmospheric and land models are Community Climate Model [CCM] 3.6 and Land Surface Model [LSM] 1.2 at T31 (~3.75° by 3.75°) resolution. These are coupled to NCAR's Ocean Model and the Community Sea Ice Model. The latter models are on a stretched grid with 0.9° equatorial and 1.8° high-latitude meridional grid spacing; zonal resolution is 3.6°. The ocean model is integrated with anisotropic horizontal viscosity (Large et al., 2001). As described in the references above, the major differences from CSM version 1 are (1) better-resolved tropical ocean circulations and thermocline structure and (2) improvements in conservation properties of flux coupler interpolation between component models.
It is not sufficient, however, to use a CGCM to model the early Paleogene ocean circulation; the model should be able to simulate tropical ocean circulations and tropical atmosphere-ocean interactions such as the El Niņo Southern Oscillation (ENSO). CSM version 1 has this capability (Large et al., 2001; Otto-Bliesner and Brady, 2001). Importantly, this version incorporates very small values of vertical diffusion and an implementation of anisotropic horizontal viscosity, providing excellent resolution of tropical current systems. The paleoposition determined during Leg 199 for the Paleogene suggests that it is the structure and properties of the thin (100-300 km), vigorous (100 cm/s), and variable eastward flowing North Equatorial Countercurrent (NECC) and North Subsurface Countercurrent (NSCC) that are likely to be critical.
Today the Pacific NECC and Equatorial Undercurrent (EUC) are, on average, vigorous eastward-flowing tropical currents, originating in the western Pacific warm pool, set up by pressure gradients associated with the across-Pacific thermocline tilt; the latitude of the NECC appears to be related to the Intertropical Convergence Zone (ITCZ) (Donguy and Meyers, 1996; Johnson et al., 2001, and references therein). These currents' magnitudes are tied to that of the Walker cell, which is related to the tilt of the thermocline (Rowe et al., 2000). The NSCC occurs at several hundred meters depth between 4° and 8° latitude and is linked to the 13°C thermostad in the eastern Pacific (Marin et al., 2000). As a consequence of their intimate ocean-atmosphere coupling, these currents change strength and depth extensively over the course of seasonal and ENSO cycles (Johnson et al., 2000). Overall, these currents usually provide a broad region of relatively fresh eastward flow at depths between 100 and 400 m, which I refer to as the NECC/NSCC complex. The dynamics of these currents were not well resolved in any previous Paleogene ocean-modeling investigation.
The details of the method developed to perform deep paleoclimate CGCM integrations are described in Huber and Sloan (2001). The topography and vegetation distributions used are described and extensively referenced in Sewall et al. (2000). Description of how the initial SST distributions were created can be found in Sloan et al. (2001). The creation of a bathymetric data set for the early Paleogene began with land-sea distributions created by Sewall et al. (2000), which were modifications to those described in Sloan and Rea (1996). Those land-sea distributions were then remapped from the 2° x 2° distribution, described in Sewall et al. (2000), to the stretched grid of the ocean model. Isochron maps created by Royer et al. (1992) were then used to identify the paleolocations of Chron 25 and other late Paleocene-early Eocene isochrons in order to constrain the location of mid-ocean ridges during the early Paleogene. Further description of the creation of the bathymetry is found in Huber et al. (in press). The solar constant, orbital parameters, and all trace gas concentrations (excluding carbon dioxide) were set at preindustrial values.
Early Paleogene climate proxies that might be used as initial conditions for a modeling experiment are sparsely distributed in time and space, have inherent uncertainties associated with them, and may not constrain quantities that are important for climate modeling (e.g., salinity). There is also a good deal of uncertainty in boundary conditions, such as topographic or bathymetric reconstructions. I have preliminarily assessed the sensitivity of my spin-up method to uncertainty in bathymetry and initial conditions using an approach called "degradation" (covered in Huber and Sloan, 2001, and Huber et al., submitted) and demonstrated that there is little sensitivity to these features. Beginning with zonally constant ocean temperatures close to the proxy estimates of Zachos et al. (1994) for the early Eocene, the simulation has been integrated for over 3200 yr in the deep ocean with the accelerated, partially asynchronous spin-up described in Huber and Sloan (2001). The simulation was then integrated in fully coupled mode for an additional 150 yr; averages over the last 50 yr are described herein. Trends in temperature and salinity were negligible (similar to values in Huber and Sloan, 2001). The tropical ocean properties produced in the simulation described here are nearly identical to those produced in the Huber and Sloan (2001) study despite the fact that the specified initial SST distributions were 4° cooler at low latitudes and 4° warmer at high latitudes, and deep ocean temperatures were 4° warmer than in the former study. This demonstrates that the modeling framework used here is robust.